Qwen2 示例代码

安装 Qwen2 示例

sudo apt install es-sdk-sample-npu-qwen

安装后 sample 位于 /opt/eswin/sample-code/npu_sample/qwen_sample 目录

使用示例

下载编译好的模型

安装 modelscope
```
pip3 install modelscope
```

下载模型

mkdir Qwen2 && cd Qwen2
modelscope download --model ZIFENG278/Qwen2-0.5B_ENNP --local_dir ./

修改 config.json 文件路径

请参考 /opt/eswin/sample-code/npu_sample/qwen_sample/src/qwen2_0_5b/config.json

{
    "model_dir": "/opt/eswin/sample-code/npu_sample/qwen_sample/models/qwen2_0_5b/",
    "block_models": [
        "modified_block_0_npu_b1.model",
        "modified_block_1_npu_b1.model",
        "modified_block_2_npu_b1.model",
        "modified_block_3_npu_b1.model",
        "modified_block_4_npu_b1.model",
        "modified_block_5_npu_b1.model",
        "modified_block_6_npu_b1.model",
        "modified_block_7_npu_b1.model",
        "modified_block_8_npu_b1.model",
        "modified_block_9_npu_b1.model",
        "modified_block_10_npu_b1.model",
        "modified_block_11_npu_b1.model",
        "modified_block_12_npu_b1.model",
        "modified_block_13_npu_b1.model",
        "modified_block_14_npu_b1.model",
        "modified_block_15_npu_b1.model",
        "modified_block_16_npu_b1.model",
        "modified_block_17_npu_b1.model",
        "modified_block_18_npu_b1.model",
        "modified_block_19_npu_b1.model",
        "modified_block_20_npu_b1.model",
        "modified_block_21_npu_b1.model",
        "modified_block_22_npu_b1.model",
        "modified_block_23_npu_b1.model"
    ],
    "logits": true,
    "logits_models": "lm_npu_b1.model",
    "tokenizer_file": "/opt/eswin/sample-code/npu_sample/qwen_sample/src/qwen2_0_5b/qwen.tiktoken",
    "vocab_size": 151936,
    "embedding_file": "/opt/eswin/sample-code/npu_sample/qwen_sample/models/qwen2_0_5b/embedding.bin",
    "embedding_size": 896,
    "num_head": 2,
    "embedding_dim": 64,
    "embedding_dim1": 4,
    "embedding_dim2": 16,
    "token_num": 1024,
    "token_n1": 64,
    "token_n2": 16,
    "precision": 2,
    "language": "en",
    "eos_list": ["<|endoftext|>", "<|im_end|>", "<|im_start|>"],
    "response_max_len": 1024,
    "response_line_len": 100,
    "prompts" : {
        "0":"介绍一下大语言模型",
        "1":"The quantum computers",
        "2":"Humans and robots coexist"
    },
    "random_cnt": 4,
    "repeat_len": 10,
    "repeat_cnt": 2,
    "log_level": 0,
    "data_dump": 0,
    "data_dump_token": [0, 180],
    "perf_enable": 0,
    "op_dump": 0,
    "op_dump_token": [0, 180],
    "op_dump_layer": [0, 23],
    "op_dump_type": 1,
    "op_dump_list": [0, 220],
    "model_test": 0,
    "repetition_penalty": 1.1,
    "top_k": 1
}

启动示例

$ sudo /opt/eswin/sample-code/npu_sample/qwen_sample/bin/es_qwen2 ./config.json

Loading models: [==================================================] 100.00% (5.46 seconds)
----------------------------------------------------------------------------------
0: 介绍一下大语言模型
1: The quantum computers
2: Humans and robots coexist
3: Customized prompts
----------------------------------------------------------------------------------
[YOU]: